The frequency shift due to atomic collisions is a major, and in some cases the dominant, limitation to the
accuracy of caesium fountain primary frequency standards. A correction for this shift is usually obtained by
measuring the frequency of the standard as a function of atomic density and performing an extrapolation to zero
density. In general this means that additional measurement time is needed to reach a given statistical resolution.
Recently, we have observed that, for a certain range of fountain parameters (i.e. the initial size of the atom
cloud and its temperature at launch), the collisional frequency shift varies significantly when the population of
the clock states (set by the first Ramsey interaction) is varied. In particular, the collisional shift can be zero for a
certain value of the population ratio. This demonstration of collisional shift cancellation offers the intriguing
prospect of operating the fountain at the zero-shift point, avoiding the need for extrapolation. In this
contribution we provide further experimental validation of the theoretical model describing the collisional shift
variation. We also discuss requirements for and benefits of the operation at the zero shift point. In addition, we
point out the possible consequences of collisional shift variation for operation of a fountain standard at elevated
microwave power, a mode of operation frequently used for the evaluation of other systematic frequency shifts.